Similar to the replication mode, in erasure coding mode, ACOS allocates data blocks and parity blocks (collectively referred to as coding blocks) based on cluster topology information, attempting to evenly distribute these blocks across different racks, chassis, and hosts to improve the overall fault tolerance of the cluster. During runtime, ACOS also automatically migrates coding blocks according to their distribution and topology information to ensure that the cluster continues to operate with a high level of fault tolerance.
In general, the more racks and chassis available in a cluster, the higher the level of fault tolerance it can provide. As long as the number of failed nodes in the cluster does not exceed the number of parity blocks M, the remaining nodes can reconstruct the data to the expected state, ensuring that services remain unaffected. Based on the number of racks and chassis, as well as the distribution of nodes across racks and chassis, the following table uses recommended erasure coding configurations (2+1, 4+1, 2+2, 4+2, and 8+2) as examples to illustrate the relationship between the number of data blocks K, the number of parity blocks M, the number of chassis B, the number of racks R, and the types of failures that can be tolerated.
| Data block count K | Parity block count M | Chassis count B | Rack count R | Tolerable fault type |
|---|---|---|---|---|
| 2 | 1 | ≤ 2 | ≤ 2 | Any single host failure |
| ≥ 3 | ≤ 2 | Any single chassis failure | ||
| ≥ R | ≥ 3 | Any single rack failure | ||
| 4 | 1 | ≤ 4 | ≤ 4 | Any single host failure |
| ≥ 5 | ≤ 4 | Any single chassis failure | ||
| ≥ R | ≥ 5 | Any single rack failure | ||
| 2 | 2 | 1 | 1 | Any two host failures |
| 2 ≤ B ≤ 3 | 1 | Any single chassis failure | ||
| ≥ 4 | 1 | Any two chassis failures | ||
| ≥ R | 2 ≤ R ≤ 3 | Any single rack failure | ||
| ≥ R | ≥ 4 | Any two rack failures | ||
| 4 | 2 | 1 ≤ B ≤ 2 | ≤ 2 | Any two host failures |
| 3 ≤ B ≤ 5 | ≤ 2 | Any single chassis failure | ||
| ≥ 6 | ≤ 2 | Any two chassis failures | ||
| ≥ R | 3 ≤ B ≤ 5 | Any single rack failure | ||
| ≥ R | ≥ 6 | Any two rack failures | ||
| 8 | 2 | 1 ≤ B ≤ 4 | ≤ 4 | Any two host failures |
| 5 ≤ B ≤ 9 | ≤ 4 | Any single chassis failure | ||
| ≥ 10 | ≤ 4 | Any two chassis failures | ||
| ≥ R | 5 ≤ R ≤ 9 | Any single rack failure | ||
| ≥ R | ≥ 10 | Any two rack failures |
